Chen Jin · Chen Jin

Hi! My name is Chen Jin, and I’m a research scientist at the Centre for Artificial Intelligence (CAI) at Data Science & Artificial Intelligence, AstraZeneca (UK). My research focuses on Large Multimodal Generative Models, Vision-Language Human-Machine Interactions, Computer Vision and Natural Language Processing for Healthcare, and Predictive Models.

Previously, I was a research assistant at the Centre for Medical Image Computing (CMIC) at UCL’s AI Centre, where I worked on Multimodal-Multiscale-Multitask representation learning, image generation and mapping. I have also developed attention-based learnable sampling methods for either data efficient or computation efficient machine learning, and applied on segmentation of large volume high-resolution images.

Externally, I got the privilege to collaborate closely with Dr Ryutaro Tanno from Google DeepMind, Prof. Daniel Alexander from CMIC UCL and Prof Mihaela van der Schaar from Artificial Intelligence and Medicine at the University of Cambridge.

I obtained my PhD in Computational Geoscience at the Institute of GeoEnergy EngineeringHeriot-Watt University. My thesis was about developing a Hierarchical Digital Rock Analysis Workflow, involving machine learning-based multi scale-modal image fusion, reconstruction and fluid simulation.

News:

Featured Research

Chen Jin, Ryutaro Tanno, Tom Diethe, Philip Teare

January 2026 Preprint

Scalable oversight via confidence-guided reasoning; enables safe self-correction matching parallel decoding accuracy at orders-of-magnitude lower compute.

CoRefine: Confidence-Guided Self-Refinement for Adaptive Test-Time Compute

Chen Jin, Ryutaro Tanno, Amrutha Saseendran, Tom Diethe, Philip Teare

January 2025 ICML, 2025

Created Lavender, an SFT method aligning VLM text-vision attention with Stable Diffusion, boosting Llama-3.2-11B and MiniCPM-v2.5 by up to 30% on 20 tasks.

Diffusion Instruction Tuning

Zhihua Liu, Amrutha Saseendran, Lei Tong, Xilin He, Fariba Yousefi, Nikolay Burlutskiy, Dino Oglic, Tom Diethe, Philip Teare, Huiyu Zhou, Chen Jin

January 2025 ICML, 2025

Training-free prompt learning for language-grounded segmentation using token-level cross-attention from a frozen diffusion model to generate object masks.

Segment Anyword

Abdullah Abdulaal, Chen Jin, Nina Montaña-Brown, Aryo Pradipta Gema, Daniel C Castro, Daniel C Alexander, Philip Teare, Tom Diethe, Dino Oglic, Amrutha Saseendran

November 2024 ICLR, 2025

A test-time LLM ensembling strategy that dynamically adapts to balance performance, diversity, and consistency, achieving state-of-the-art results.

Dynamic Mixture of Agents (DMoA)

Ana-Paula Gema, Chen Jin, Abdullah Abdulaal, Tom Diethe, Philip Teare, Benjamin Alex, Pasquale Minervini, Amrutha Saseendran

October 2024 EMNLP, 2025

Training-free decoding that mitigates LLM hallucinations by contrasting a base model with a masked-retrieval variant, boosting summarisation by up to 18.6%.

Decoding by Contrasting Retrieval Heads

Chen Jin, Ryutaro Tanno, Amrutha Saseendran, Tom Diethe, Philip Teare

May 2024 ICML, 2024

Personalised image generation extending textual inversion for mask-free learning of multiple concepts from a single sentence–image pair (Stable Diffusion).

An Image is Worth Multiple Words: Discovering Object Level Concepts using Multi-Concept Prompt Learning

Seunghoi Kim, Chen Jin, Tom Diethe, Matteo Figini, Henry F. J. Tregidgo, Asher Mullokandov, Philip Teare, Daniel C. Alexander

April 2024 ECCV, 2024, Oral

Training-free diffusion framework that reduces hallucinations via multiple local diffusion processes, cutting hallucinations by 40% (medical) and 25% (natural).

Tackling Structural Hallucination in Image Translation with Local Diffusion

Chen Jin, Ryutaro Tanno, Thomy Mertzanidou, Eleftheria Panagiotaki, Daniel C. Alexander

January 2022 ICLR, 2022

We introduce a learned, adaptive downsampling method that prioritizes challenging regions, enabling efficient segmentation of high-res images on limited computing.

Learning to Downsample for Segmentation of Ultra-High Resolution Images

Chen Jin, Ryutaro Tanno, Moucheng Xu, Thomy Mertzanidou, Daniel C. Alexander

July 2020 arXiv, 2020

Building on our prior foveation work (MICCAI 2020), we introduce a more computationally efficient hard-gated categorical sampling method for FoV-resolution patch configurations with two differentiable solutions. We …

Foveation for Segmentation of Ultra-High Resolution Images

Chen Jin, Ryutaro Tanno, Moucheng Xu, Thomy Mertzanidou, Daniel C. Alexander

April 2020 MICCAI, 2020

We introduce a foveation module that dynamically adjusts patch FoV and resolution for ultra-high resolution image segmentation, achieving state-of-the-art results and significant accuracy boosts on challenging datasets.

Foveation for Segmentation of Mega-Pixel Histology Images

Research Outputs

Explore the full publication list or jump to my Google Scholar profile.

*

Zhihua Liu, Amrutha Saseendran, Lei Tong, Xilin He, Fariba Yousefi, Nikolay Burlutskiy, Dino Oglic, Tom Diethe, Philip Teare, Huiyu Zhou, Chen Jin

January 2025 ICML, 2025

Segment Anyword

Training-free prompt learning for language-grounded segmentation using token-level cross-attention from a frozen diffusion model to generate object masks.

Abdullah Abdulaal, Chen Jin, Nina Montaña-Brown, Aryo Pradipta Gema, Daniel C Castro, Daniel C Alexander, Philip Teare, Tom Diethe, Dino Oglic, Amrutha Saseendran

November 2024 ICLR, 2025

Dynamic Mixture of Agents (DMoA)

A test-time LLM ensembling strategy that dynamically adapts to balance performance, diversity, and consistency, achieving state-of-the-art results.

Seunghoi Kim, Chen Jin, Tom Diethe, Matteo Figini, Henry F. J. Tregidgo, Asher Mullokandov, Philip Teare, Daniel C. Alexander

April 2024 ECCV, 2024, Oral

Tackling Structural Hallucination in Image Translation with Local Diffusion

Training-free diffusion framework that reduces hallucinations via multiple local diffusion processes, cutting hallucinations by 40% (medical) and 25% (natural).

Recent Posts

Contact

Read the original on chenjin.netlify.app ↗